OpenAI Chief Scientist publishes essay on alien intelligence and alignment challenges

OpenAI Chief Scientist Jakub Pachocki published “An Alien Mind,” arguing that scaled deep-learning systems produce intelligence that is not directly comparable to human intelligence and that alignment, particularly value alignment and robust generalization, remains a core unsolved challenge. The essay discusses diminishing reliability of chain-of-thought monitoring, the risks of recursive self-improvement, and the need for broader interventions beyond technical solutions while noting GPT-6 Astra as better aligned than prior models yet still requiring substantial further progress.